Survey
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
* Your assessment is very important for improving the workof artificial intelligence, which forms the content of this project
Basis (linear algebra) wikipedia , lookup
Homogeneous coordinates wikipedia , lookup
Factorization of polynomials over finite fields wikipedia , lookup
Congruence lattice problem wikipedia , lookup
Eisenstein's criterion wikipedia , lookup
Covering space wikipedia , lookup
Lines on Projective Hypersurfaces Roya Beheshti Abstract We study the Hilbert scheme of lines on hypersurfaces in the projective space. The main result is that for a smooth Fano hypersurface of degree at most 6 over an algebraically closed field of characteristic zero, the Hilbert scheme of lines has always the expected dimension. 1 Introduction Let k be an algebraically closed field and X ⊂ Pnk a projective variety. Denote by F (X) the Hilbert scheme of lines on X. It is a subscheme of the Grassmannian of lines in Pnk and is called the Fano variety of lines on X. We are interested in studying F (X) when X is a hypersurface. For general hypersurfaces, these schemes have been studied classically, but little is known when X is not general. It is known that for a general hypersurface X ⊂ Pnk of degree d, F (X) is smooth and has the expected dimension 2n − d − 3. It is also known that F (X) may be reducible or non-reduced for particular hypersurfaces, even smooth ones. What can be said about dimension of F (X) when X is an arbitrary smooth hypersurface? O. Debarre and J. de Jong, independently, asked the following question in this regard. Question 1.1. Let k be an algebraically closed field of characteristic p, and let X ⊂ Pnk be a hypersurface of degree d. Assume d ≤ n, and if p > 0, assume furthermore that d ≤ p. Does F (X) have the expected dimension 2n − d − 3 whenever X is smooth? When p is positive and d ≥ p + 1, F (X) does not always have the expected dimension. It is easy to see that the family !n of lines contained in the Fermat hypersurface of degree p + 1 in Pnk with equation i=0 xp+1 = 0 has dimension at least 2n − 6, which i is larger than the expected dimension when p is odd ([4], 2.15). Also, it can be shown that if char(k) = 0 and X is the Fermat hypersurface of degree d ≥ n in Pnk , then dim F (X) = n − 3, which is larger than the expected dimension for d > n ([4], Exercise 2.5). Hence the assumptions in Question 1.1 are necessary. In [7], Harris et al. gave a positive answer to the above question when p = 0 and d is very small with respect to n. The main result of this paper is the following. Theorem 4.2. Assume k has characteristic zero and X is any smooth hypersurface of degree d ≤ 6 in Pnk . Then F (X) has the expected dimension when d ≤ n. 1 The case d = 3 of the above theorem is elementary. The case d = 4 is due to A. Collino; he proved in [3] that Question 1.1 holds true for all smooth quartic hypersurfaces when the characteristic of the base field is not 2 or 3. Also, the case d = 5 of the above theorem was proved by O. Debarre before, but our approach here is different from the previous ones and allows us to treat all cases d ≤ 6 in a unified way. This paper is organized as follows. The proof of the above theorem is given in Section 4. The results of Sections 2 and 3 are necessary for the proof of this theorem. In Section 2, we show that a subvariety of F (X) that sweeps out a divisor in X cannot be uniruled when d ≥ n − 1. In Section 3, we discuss a theorem of J. Landsberg on the dimension of the family of lines having contact to a specific order with a hypersurface at a general point, and we use the same method used in the proof of his theorem to prove a proposition on the singularities of the second fundamental forms of hypersurfaces. 1.1 Conventions. 1. All schemes are considered over a fixed algebraically closed field of characteristic zero, and all points are closed points unless otherwise stated. 2. For any projective variety X ⊂ Pn , F (X) always denotes the Fano variety of lines on X. 3. For a scheme X and a sheaf F of OX -modules on X, we denote by F /tor the sheaf obtained from dividing out F by its torsion. 4. If X is a scheme and Y is a closed subscheme of X with ideal sheaf I , then we denote by NY /X the normal sheaf of Y in X: NY /X = Hom(I /I 2 , OY ). Acknowledgment. The results in this paper form part of my Ph.D. thesis at the Massachusetts Institute of Technology. I am extremely grateful to my thesis advisor, Johan de Jong, for his enumerable insights and helpful suggestions. I also thank Jason Starr for patiently answering my questions, and the referee for many useful suggestions in improving this manuscript. 2 Rational curves on the Fano variety of lines on complete intersections A projective variety Y of dimension m is called uniruled if there exist a variety Z of dimension m − 1 and a dominant rational map P1 × Z !!" Y . Let X ⊂ Pn be a smooth complete intersection of type (d1 , . . . , dl ), and let d = !l i=1 di . The purpose of this section is to prove the following theorem. Theorem 2.1. Let Y be an irreducible closed subvariety of F (X) such that the lines corresponding to its points cover a divisor in X. If d ≥ n − 1, then Y is not uniruled. 2 The assumption d ≥ n − 1 is necessary: let X be the Fermat hypersurface of degree d ≤ n − 2 in Pn given by the equation xd0 + xd1 + · · · + xdn = 0, and let p be the point with coordinates (0; . . . ; 0; a; b) in X. Then the lines on X passing through p form a cone over the hypersurface xd0 + · · · + xdn−2 = 0 in Pn−2 , and this hypersurface is uniruled, so we get a uniruled family of lines on X which sweeps out a divisor. Remark 2.2. Let X be a smooth hypersurface. If d = n and X is general, then F (X) is irreducible and the normal bundle of a general line l on X is isomorphic to Oln−3 ⊕ Ol (−1) ([9], V.4.4). Hence the lines on X sweep out a divisor and from the theorem above, we can conclude that F (X) is not uniruled. If d = n − 1, then X is always covered by lines and hence there are many such Y. If d > n and X is general, then the lines on X sweep out a subvariety of codimension at least 2, but for special hypersurfaces, it is possible that they cover a divisor in X; this happens for example in the case of Fermat hypersurfaces. Proof of Theorem 2.1. Assume on the contrary that Y is covered by rational curves, and let m = dim X = n − l. Without loss of generality, we can assume that our base field is uncountable. We define Σe to be the ruled surface P(OP1 ⊕ OP1 (−e)) over P1 . Let C be a rational curve on F (X), and denote by S ⊂ C × X the family of lines parametrized by C. Let ν : P1 → C be the normalization. The ruled surface S ×C P1 is isomorphic to Σe for some nonnegative integer e, and we have the following diagram f : Σe " S ⊂C ×X ! P1 "X ! " C. Let Mor(Σe , X) denote the scheme parametrizing morphisms from Σe to X. For every integer e, Mor(Σe , X) has countably many irreducible components. Since our base field is uncountable and the lines corresponding to the points of Y cover a subvariety of codimension 1 in X, there is a nonnegative integer e0 and an irreducible subvariety of Mor(Σe0 , X), denoted by Z, with the property that points of Z correspond to rational curves on Y, and the image of the map φ : Z × Σe0 ([g], p) " X ! " g(p) is at least (m − 1)-dimensional. For the rest of the argument, we put e = e0 and Σ = Σe0 , and we let [f ] denote a general point in Z. There is an injective morphism from the tangent sheaf of Σ to the pullback of the tangent sheaf of X. Denote the quotient by G 3 0 → TΣ → f ∗ TX → G → 0. (1) "m−3 To get a contradiction, we compute h0 (Σ, ( G )/tor) in two different ways. First, "m−3 we use the fact that deformations of f cover a divisor in X to show that h0 (Σ, ( G )/tor) is positive and then, we prove that it is zero, using some computations with exact sequences of powers of tangent sheaves. "m−3 Step 1: h0 (Σ, ( G )/tor) > 0. Let p be a general point of Σ and consider the evaluation map and the tangent map α : H 0 (Σ, f ∗ TX ) −→ (f ∗ TX )p ∼ = TX,f (p) , β : TΣ,p −→ TX,f (p) to f at p. Lemma 2.3. For a general point p of Σ the image of the map α⊕β H 0 (Σ, f ∗ TX ) ⊕ TΣ,p −−−−→ TX,f (p) is at least (m − 1)-dimensional. Proof. Since the image of φ is at least (m − 1)-dimensional, the same is true for the image of the differential map d([f ],p) φ at the general point ([f ], p). The Zariski tangent space to Z at [f ] is a subspace of the Zariski tangent space to Mor(Σ, X) at [f ] which is isomorphic to H 0 (Σ, f ∗ TX ) ([9], I.2.8), and d([f ],p) φ is the restriction of α ⊕ β to this subspace. This proves the lemma. Look at the following commutative diagram. H 0 (Σ, TΣ ) " H 0 (Σ, f ∗ TX ) ! ! α TΣ,p β " TX,f (p) . For a general point p of Σ, the map H 0 (Σ, TΣ ) → TΣ,p is surjective. Therefore, for such a point, β(TΣ,p ) ⊂ α(H 0 (Σ, f ∗ TX )). So by Lemma 2.3, the image of α is at least (m−1)-dimensional. Consider now sequence (1). We have shown that the image of α is at least (m − 1)-dimensional. This implies that the global sections of f ∗ TX generate a subspace of dimension at least m − 1 at a general point of Σ. Therefore global sections of G generate "m−3 a subspace of dimension at least m − 3 at a general point of Σ and hence h0 (Σ, ( G )/tor) > 0. 4 "m−3 Step 2: h0 (Σ, ( G )/tor) = 0. From sequence (1) we get a morphism "2 "m−3 TΣ ⊗ ( G )/tor ξ1 ∧ ξ2 ⊗ η̄1 ∧ · · · ∧ η̄m−3 ! " "m−1 f ∗ T X " β(ξ1 ) ∧ β(ξ2 ) ∧ η1 ∧ · · · ∧ ηm−3 , where ηi is any lifting of η̄i in sequence (1). Furthermore, this map is injective since it "2 "m−3 is injective at the generic point of Σ and TΣ ⊗ ( G )/tor is torsion free. If we twist the above map with the canonical sheaf ωΣ of Σ, we get another injective morphism #m−1 #m−3 G )/tor −→ f ∗ TX ⊗ ωΣ . 0 −→ ( "m−1 ∗ We compute h0 (Σ, f TX ⊗ ωΣ ) and show that it is zero. This will conclude the proof of Theorem 2.1. We have h0 (Σ, #m−1 f ∗ TX ⊗ ωΣ ) = h2 (Σ, #m−1 ∨ f ∗ TX ) ∨ = h2 (Σ, f ∗ TX ⊗ det f ∗ TX ) = h2 (Σ, f ∗ TX ⊗ OX (d − n − 1)). Pulling back the sequence of tangent sheaves, we get the following exact sequence 0 → f ∗ TX → f ∗ TPn |X → $ i f ∗ OX (di ) → 0. (2) ∨ Twisting the above sequence with det f ∗ TX and applying the long exact sequence of cohomology, we see that to prove the assertion, it suffices to prove two things: (a) h1 (Σ, f ∗ OX (di + d − n − 1)) = 0, for 1 ≤ i ≤ l. (b) h2 (Σ, f ∗ TPn |X ⊗ f ∗ OX (d − n − 1)) = 0. Let F be the class of a fiber of π : Σ → P1 and E the class of the section with E = −e. Recall that the Picard group of Σ is the free abelian group generated by F and E, and the intersection products are given by 2 E 2 = −e, F 2 = 0, E · F = 1. Let f ∗ OX (1) = aE + bF. The image of a line of ruling of Σ under f is a line, so 1 = f ∗ OX (1) · F = (aE + bF ) · F = a. Also f ∗ OX (1) · E ≥ 0, so 5 −e + b = (E + bF ) · E ≥ 0. Let d%i = di + d − n − 1. We have f ∗ OX (d%i ) = d%i E + bd%i F. Since the first cohomology group of the sheaf f ∗ OX (d%i ) restricted to a fiber of π vanishes, we have R1 π∗ (f ∗ OX (d%i )) = 0, so h1 (Σ, f ∗ OX (d%i )) = h1 (P1 , π∗ f ∗ OX (d%i )) = h1 (P1 , OP1 (bd%i ) ⊗ π∗ (OΣ (d%i E))) = h1 (P1 , OP1 (bd%i ) ⊗ (OP1 ⊕ OP1 (−e) ⊕ · · · ⊕ OP1 (−ed%i ))) =0 (since b ≥ e). This proves (a). If we pullback the dual of the Euler sequence of Pn and twist it with f ∗ OX (d−n−1), we get the following exact sequence: 0 → f ∗ OX (d − n − 1) → f ∗ OX (d − n)n+1 → f ∗ TPn |X ⊗ f ∗ OX (d − n − 1) → 0. (3) So to prove (b), it is enough to show that: (b% ) h2 (Σ, f ∗ OX (d − n)) = h0 (Σ, f ∗ OX (n − d) ⊗ ωΣ ) = 0. To show that (b% ) is true, we observe that since d ≥ n − 1, the invertible sheaf f ∗ OX (n − d) ⊗ ωΣ has negative intersection with the divisor F and hence has no nonzero global sections. This completes the proof of Theorem 2.1. 3 Lines having contact to a specific order with a hypersurface Let X ⊂ Pn be a hypersurface of degree d. Take a system of homogeneous coordinates (x0 ; x1 ; . . . ; xn ) on Pn , and let P be a homogeneous polynomial of degree d vanishing on X. Fix a smooth point p in X. For 1 ≤ k ≤ d, let Ypk be the degree k hypersurface in Pn given by the homogeneous polynomial % (i1 ,...ik ) 0≤i1 ,...,ik ≤n ∂kP (p) xi1 . . . xik . ∂xi1 . . . ∂xik (4) Note that Yp1 is just the embedded tangent space to X at p. If k = 2, then the restriction of (4) to the tangent space TX,p gives a quadric form which is called the second fundamental form of X at p. Denote by Σkp the scheme theoretic intersection of Yp1 , Yp2 , . . . , Ypk . Since for a point q = (q0 ; . . . ; qn ) in Pn we have 6 P (p + λq) = d % λk k=0 k! % (i1 ,...,ik ) ∂kP (p) qi1 . . . qik , ∂xi1 . . . ∂xik 0≤i1 ,...,ik ≤n Σkp is a cone with vertex p whose underlying space is the union of all lines in Pn passing through p and having contact to order k with X at p, and Σdp is exactly the cone of lines on X passing through p. Since Σkp is the intersection of k hypersurfaces in Pn , its expected dimension is n − k. The next theorem says that generally the expected dimension is obtained unless the whole cone lies in X. Theorem 3.1 (Landsberg [10]). Let X ⊂ Pn be a hypersurface, and let p be a general point of X. Any irreducible component of Σkp which has dimension greater than n − k is contained in X. We will use this theorem several times in the proof of our main theorem. For later use, we rephrase the above theorem in the following corollary. Corollary 3.2. Let X ⊂ Pn be a projective variety of dimension m, and let p be a general point of X. Denote by Σp the subvariety of X swept out by lines passing through p, and denote by r the dimension of Σp . Then Σp has degree at most (m + 1 − r)! and it is contained in the proper intersection of the tangent plane to X at p and m − r hypersurfaces of degrees 2, 3, . . . , m + 1 − r. Proof. Assume first that X is a hypersurface in Pn . By Theorem 3.1, all the components of Σn−r are r dimensional, since otherwise Σp would be at least r +1 dimensional. p Hence the hypersurfaces Yp1 , . . . , Ypn−r intersect properly. The cone Σp is a component of this intersection and its degree is at most (n − r)!. If X has codimension greater than 1 in Pn , then we consider a general projection of X into Pm+1 . Since the projection map is generically finite, we get a hypersurface in Pm+1 such that the lines passing through its general point sweep out a subvariety of dimension r and we are in the situation of the previous case. In the next subsection, we recall some facts about frame bundles of hypersurfaces. We then prove a proposition on the singularities of the second fundamental form of a hypersurface at a general point. The proposition will be used later in the proof of Theorem 4.2. 3.1 Preliminaries on frame bundles. For a given integer n, let GLn+1 be the space of invertible matrices of size n + 1 over the base field. Let φ : GLn+1 → Pn be the map given by φ([V0 , . . . , Vn ]) = V0 , where [V0 , . . . , Vn ] is the matrix with columns V0 , . . . , Vn . Let Ω be the matrix of global 1-forms on GLn+1 given by 7 Ωf = f −1 df for f ∈ GLn+1 . Denote the entries of Ω by ωij , and let xij be the regular function on GLn+1 defined by xij (f ) = fij for 0 ≤ i, j ≤ n. We have dxij (f ) = % 0≤k≤n fik ωkj (f ) for f = (fij )0≤i,j≤n ∈ GLn+1 . (5) Notation. In what follows, f = (fij )0≤i,j≤n is always an element of GLn+1 . We denote the columns of f by V0 , . . . , Vn . Hence the i-th entry of Vj is fij . Let X be an integral hypersurface of degree d in Pn given by the homogeneous polynomial P . For given integers 0 ≤ i1 , i2 , . . . , im ≤ n, the regular function rim1 ,...,im on GLn+1 is defined by rim1 ,...,im (f ) = ∂mP +i . . . ∂ V +i ∂V 1 m (V0 ), f = [V0 , . . . , Vn ]. More precisely, % rim1 ,...,im (f ) = fj1 ,i1 . . . fjm ,im (j1 ,...,jm ) ∂mP (V0 ). ∂xj1 . . . ∂xjm (6) 0≤j1 ,...,jm ≤n Since the ωij form a basis for the space of 1–forms at every point of GLn+1 , we can express the derivative of the functions rim1 ,...,im as linear combinations of the ωij . The next lemma says what the coefficients of the combination are when i1 = · · · = im . Lemma 3.3. For 0 ≤ i ≤ n, we have % % m+1 m m dri,i,...,i = ri,...,i,t ωt0 + m ri,...,i,t ωti . 0≤t≤n 0≤t≤n Proof. By partial derivation, we have % m dri,...,i = % (fj1 ,i . . . fjˆl ,i . . . fjm ,i (j1 ,...,jm ) 1≤l≤m + % ∂mP (V0 )) dxjl ,i ∂xj1 . . . ∂xjm fj1 ,i . . . fjm ,i d( (j1 ,...,jm ) ∂mP (V0 )). ∂xj1 . . . ∂xjm Hence the assertion follows from the equality d( % ∂mP ∂ m+1 P (V0 )) = (V0 ) dxt0 , ∂xj1 . . . ∂xjm ∂xj1 . . . ∂xjm ∂xt 0≤t≤n and equation (5). 8 Before going on, we need another easy lemma. Lemma 3.4. For 0 ≤ i1 , . . . , im ≤ n, we have rim+1 = (d − m)rim1 ,...im , where 1 ,...,im ,0 d = deg X. Proof. By definition, we have rim+1 = 1 ,...,im ,0 % (j1 ,...,jm ) = (d − m) % ∂ ∂mP (V0 )) fj1 ,i1 . . . fjm ,im ( ft0 ∂xt ∂xj1 . . . ∂xjm t % fj1 ,i1 . . . fjm ,im (j1 ,...,jm ) ∂mP (V0 ) ∂xj1 . . . ∂xjm = (d − m)rim1 ,...,im . Let FX be the set of all invertible matrices with columns [V0 , . . . , Vn ] such that V0 is a smooth point of the hypersurface X and V1 , . . . Vn−1 are in the embedded tangent space to X at V0 , namely the hyperplane defined by the linear polynomial % ∂P ( (V0 )) xt . ∂xt t The scheme FX is a smooth locally closed subvariety of GLn+1 ; it is a principal bundle over the smooth locus of X and is called the frame bundle of X. Denote by νij the image of ωij under the restriction map H 0 (GLn+1 , Ω1GLn+1 ) → 0 H (FX , Ω1FX ). Lemma 3.5. At every point of FX , the following hold. (a) νn0 = 0. (b) ν00 , ν10 , . . . , νn−1,0 are independent 1-forms. Proof. (a) We have r0 = % t ft0 ∂P (V0 ) = d P (V0 ) = 0, ∂xt so dr0 = 0. Therefore, it follows from Lemma 3.3 that % % 0 = dr0 = r0t νt0 + rt νt0 . 0≤t≤n (7) 0≤t≤n Since V1 , . . . , Vn−1 are in the tangent space to X at V0 , we have r1 = · · · = rn−1 = 0. Applying Lemma 3.4, we get the equalities r0t = rt = 0 hence for 0 ≤ t ≤ n − 1, 2rn νn0 = 0. 9 Because all the matrices in FX are invertible, Vn is not in the tangent hyperplane at V0 and so rn ,= 0 and νn0 = 0. To prove part (b), observe that the map φ : GLn+1 → Pn factors through the map % φ : GLn+1 → An+1 and dxt0 is the pullback of dxt under φ% . Let X % ⊂ An+1 be the % affine cone over X. The map φ% |FX : FX −→ Xsmooth is surjective and its fibers are linear, so it is a smooth map. We know that dx0 , . . . , dxn form a space of dimension n at every smooth point of X % , hence the pullbacks of these 1-forms make an n-dimensional space at every point of FX . On the other hand, f is invertible and f Ω = df . Therefore ν00 , ν10 , . . . , νn−1,0 , νn0 form an n-dimensional space at every point and by part (a), νn0 = 0. So the first n 1-forms are linearly independent. For a point p in X, let Ypk and Σkp be defined as in the beginning of this section. Lemma 3.6. For f = [V0 , . . . , Vn ] ∈ FX , the following hold. 2 k (a) Vi ∈ ΣkV0 if and only if ri1 (f ) = ri,i (f ) = · · · = ri,i,...,i (f ) = 0. (b) Assume that V1 ∈ ΣkV0 . Then Vi is in the Zariski tangent space to ΣkV0 at V1 if and only if 2 k (f ) = · · · = r1,1,...,1,i (f ) = 0. ri1 (f ) = r1,i Proof. Part (a) follows from the definition. We prove part (b) in the case of k = 2. The proof in the general case is similar. The point Vi is in the Zariski tangent space to Σ2V0 at V1 if and only if it is in the Zariski tangent spaces to the hypersurfaces YV10 and YV20 at V1 . The equations of these tangent spaces are given by and % ∂P ( (V0 )) xt = 0, ∂xt t ) % & ∂ ' % ∂2P ( (V0 )xm xj (V1 ) xt ∂xt m,j ∂xm ∂xj t =2 % t,m ∂2P (V0 )fm1 xt . ∂xt ∂xm If we evaluate these two linear polynomials at Vi , we get ri (f ) and r1,i (f ). 10 3.2 Singularities of the second fundamental form. Let X ⊂ Pn be a hypersurface, and for a smooth point p in X, denote by Zpk the intersection of Ypk and the embedded tangent space Yp1 to X at p. In this subsection, we prove the following proposition on the singularities of Zp2 when p is general. Proposition 3.7. For a general point p of X, the singular points of Zp2 are singular points of Zpk for 2 ≤ k ≤ d = deg X.1 To prove the proposition, we need the following lemma. Lemma 3.8. For a point f = [V0 , . . . , Vn ] ∈ FX , Vj is a singular point of ZVk0 if and only if k k k (f ) = rj,...,j,1 (f ) = · · · = rj,...,j,n−1 (f ) = 0. rj,...,j,0 Proof. We can assume Yp1 is given by xn = 0. The lemma then follows easily from the definitions. Proof of Proposition 3.7. Since Zp2 is a quadric, the singular points of Zp2 form a linear subvariety. By ([6], 2.6), this linear subvariety is contained in X and is a fiber of the Gauss map. Let s be the dimension of the singular locus of Zp2 . We restrict our functions to those matrices f = [V0 , . . . , Vn ] in FX such that V1 , . . . Vs are singular points of ZV20 . So let HX ⊂ FX be the set of those matrices such that V0 is a general point of X and V1 , V2 , . . . , Vs are singular points of ZV20 , and %% let νij be the image of νij under the restriction map H 0 (FX , Ω1FX ) → H 0 (HX , Ω1HX ). Since V1 , V2 , . . . , Vs are in X, it follows from Lemma 3.6 that k =0 rj,...,j 1 ≤ j ≤ s, 1 ≤ k ≤ d, therefore 0 = drj = % 0≤t≤n−1 %% rjt νt0 + % %% rt νtj j = 1, . . . , s. (8) 0≤t≤n Since V1 , . . . , Vn−1 are in the tangent space to X at V0 , we have r1 = r2 = · · · = rn−1 = 0. The matrix f is invertible, so Vn is not contained in the tangent space to X at V0 and rn ,= 0. Also, by the last lemma, we have rj0 = · · · = rj,n−1 = 0, so equation (8) %% implies that νnj = 0. Applying Lemma 3.3, we get 1 J. Landsberg mentioned to us that this proposition also follows from [11, Thm 3.1]. 11 0 = drjj n n % % 3 %% 2 %% = rj,j,t νt0 +2 rj,t νtj t=0 = n−1 % t=0 3 %% rj,j,t νt0 . t=0 %% %% , . . . νn−1,0 are linearly independent [Notice that in Lemma By Lemma 3.5, the forms ν00 3.5, we proved the independence of these forms on FX . The same proof works here by generic smoothness and the fact that the fibers of φ% |HX are linear.], hence rj,j,0 = · · · = rj,j,n−1 = 0. This implies that Vj is a singular point of Zp3 . By repeating this argument, we see that V1 , . . . , Vs are singular points of all the Zpk . 4 Dimension of the Fano variety of lines on hypersurfaces Let X ⊂ Pn be a hypersurface of degree d given by a homogeneous polynomial P . Denote by G(1, n) the Grassmannian which parametrizes lines in Pn . Since X is a hypersurface, it is easy to describe F (X) as a subscheme of G(1, n). Let S be the n+1 universal rank 2 subbundle of OG(1,n) . The restriction of S to any point [l] in G(1, n) is identified with the rank 2 linear subspace of An+1 whose projective space is l ⊂ Pn . Hence P gives rise to a section of Symd (S ∨ ), and the scheme theoretic zero locus of this section is exactly F (X). Therefore the ideal sheaf of F (X) is locally generated by d + 1 elements and if the corresponding global section of Symd (S ∨ ) is regular, then the dimension of F (X) is dim G(1, n) − (d + 1) = 2n − d − 3. We refer to the number 2n − d − 3 as the expected dimension of F (X). This description of F (X) shows that the dimension of F (X) is always greater than or equal to 2n − d − 3. The following well-known lemma asserts that for a general hypersurface X, the scheme F (X) has the expected dimension. Lemma 4.1. For every hypersurface X of degree d in Pn , dim F (X) ≥ 2n − d − 3. For a general X, F (X) has dimension 2n − d − 3 if d ≤ 2n − 3 and is empty otherwise. This result can be obtained by calculating the dimension of the tangent space to F (X) at a general point. For a proof see [9, Thm V.4.3] or [5] where a more general statement on the space of linear subvarieties of complete intersections is given. In this section we prove the following theorem. Theorem 4.2. If X ⊂ Pn is any smooth Fano hypersurface of degree d ≤ 6, then F (X) has the expected dimension 2n − d − 3. 12 We should remark that for special smooth hypersurfaces, F (X) might be nowhere reduced and so the dimension of the Zariski tangent space to F (X) might be larger than the expected dimension of F (X) at every point. For example, if X is the Fermat hypersurface of degree 4 in P4 , then F (X) has 40 1-dimensional components each with multiplicity 2 (see [4], Exercise 2.5). This example also shows that F (X) can be reducible if X is not general. However, if d ≤ 2n − 4 and X is not a quadric surface, then F (X) is always connected even if X is not smooth (see [9, Thm V.4.3] or [1]). Reduction to the case of n = d. Before giving the proof of the theorem, we show that to prove Question 1.1 holds true for a given degree d, it is enough to consider only the case n = d. Lemma 4.3. If Question 1.1 is true for d = n, then it is true for d ≤ n. Proof. Fix an integer d. We show that if d ≤ m and if the statement of the question holds for every smooth hypersurface of degree d in Pm , then it holds for every smooth hypersurface of degree d in Pm+1 . Let X be a smooth hypersurface of degree d in Pm+1 , and let X % be a general hyperplane section of X. Since X % is smooth, by our assumption, dim F (X % ) = 2m − d − 3. Let Y be an irreducible component of F (X). By the following lemma, either a codimension at most 2 subvariety of Y lies in F (X % ) or all the lines corresponding to the points of Y pass through the same point x. In the former case, we get dim Y ≤ dim F (X % ) + 2 = 2(m + 1) − d − 3. In the latter case, all these lines are contained in the intersection of X and the tangent hyperplane at x. Hence dim Y ≤ m − 1 and the equality holds if and only if X is a hyperplane, therefore dim Y ≤ m − 2 ≤ 2(m + 1) − d − 3. Lemma 4.4. Let Y be an irreducible subvariety of G(1, n). If there is a hyperplane Λ ⊂ Pn such that the intersection of Y and the family of lines in Λ has codimension greater than 2 in Y, then all the lines corresponding to the points of Y pass through the same point. Proof. Let dim Y = s. Notice that by our hypothesis, s ≥ 3, so the lines corresponding to the points of Y cannot all lie on the same plane. Therefore if we show that every two lines of Y intersect, we can conclude that all of them pass through the same point. Let I ⊂ (Pn )∗ × Y be the incidence correspondence, and let p and q be the projections from I to (Pn )∗ and Y respectively. Every fiber of q is linear of dimension n − 2, hence I is irreducible of dimension s + n − 2. By our assumption, there is a fiber of p whose dimension is at most s − 3, so p is not dominant and therefore any non-empty fiber of p has dimension at least s + n − 2 − (n − 1) = s − 1. For a point [l] ∈ Y, let Λl ⊂ (Pn )∗ be the set of hyperplanes which contain l. Two lines l and l% in Pn intersect if and only if dim(Λl ∩ Λl! ) = n − 3. Let Il ⊂ Λl × Y be the incidence correspondence, and let pl and ql be the projections from Il to Λl and Y. We have shown that any fiber of pl has dimension at least s − 1. Thus dim Il ≥ s − 1 + dim Λl = s + n − 3. So the dimension of any fiber of ql is at least s + n − 3 − dim Y = n − 3. 13 4.1 Proof of Theorem 4.2 for d ≤ 5. Lemma 4.5. A smooth hypersurface of degree d in Pn is not covered by lines if d ≥ n. Proof. Let X ⊂ Pn be a smooth hypersurface which is covered by lines. Let I ⊂ X × F (X) be the universal family of lines on X, and let (q, [l]) be a general point of I. Consider the following commutative diagram: TI,(q,[l]) " TF (X),[l] ∼ = H 0 (l, Nl/X ) dp1 ! TX,q φ ψ " Nl/X ! ⊗ κ(q), where p1 is the projection map from I to X, dp1 is its differential, and ψ is the map induced by the natural map from the tangent sheaf of X to the normal sheaf of l in X. Let Nl/X ∼ = Ol (a1 ) ⊕ · · · ⊕ Ol (an−2 ) be the decomposition of Nl/X into line bundles. Since X is covered by lines, the dimension of the image of dp1 is at least n − 1, so the dimension of the image of ψ◦dp1 is at least n−2. Therefore dim φ(H 0 (l, Nl/X )) ≥ n−2. This implies that each ai should be non-negative. On the other hand, by the following exact sequence, a1 + · · · + an−2 = n − 1 − d, 0 → Nl/X → Nl/Pn ∼ = Ol (1)n−1 → NX/Pn |l ∼ = Ol (d) → 0 thus n − 1 − d ≥ 0. We are now ready to prove Theorem 4.2 for d ≤ 5. n = d = 3. The lemma above shows that X is not covered by lines, so there are only finitely many lines on X. n = d = 4. By the lemma above, the union of all lines on X form a subvariety of dimension at most 2. Any 2-dimensional subvariety of P4 which contains a 2-parameter family of lines is a linear subvariety. Since X is smooth, it cannot contain a plane: if X contains a plane, then we can assume that the plane is given by x0 = x1 = 0, so X is defined by an equation of the form x0 P0 + x1 P1 , where P0 and P1 are degree 3 polynomials; this is not possible since any point in the intersection x0 = x1 = P0 = P1 = 0 would be a singular point of X. Hence dim F (X) = 1. n = d = 5. Assume on the contrary that dim F (X) ≥ 3. Let Y be an irreducible subvariety of F (X) whose dimension is 3, and let X % be the subvariety of X swept out by the lines corresponding to the points of Y. By Lemma 4.5, dim X % ≤ 3, and since a surface can contain at most a 2-parameter family of lines, X % is 3-dimensional. Since X % contains a 3-parameter family of lines, there is at least a 1-parameter family of lines passing through its general point. Corollary 3.2 yields that the degree of the cone of such lines is at most 2, hence it is a cone over a rational curve. This contradicts Theorem 2.1. 14 Remark 4.6. In [2], an alternative proof of the case n = d = 5 is given. It is shown, by computing the derivative of the Abel-Jacobi map, that if a smooth quintic threefold contains a 1-parameter family of lines, then its Abel-Jacobi map is nonzero. Hence such a hypersurface cannot be a general hyperplane section of a smooth hypersurface of degree 5 in P5 . 4.2 Proof of Theorem 4.2 for d = 6. Outline of proof. Without loss of generality, we can assume that our base field is uncountable, and by Lemma 4.3, it is enough to consider the case n = d = 6. Assume on the contrary that X is a smooth hypersurface of degree 6 in P6 such that dim F (X) > 2n − d − 3 = 3. Let X % be the subvariety of X swept out by its lines. We show that our assumptions imply that X % has codimension 1 in X and the dimension of the family of lines on X % passing through a general point is 1. Let Σ be the cone of lines passing through a general point of X % , and let C be a general hyperplane section of Σ. We use the results of Sections 2 and 3 to show that C is a non-rational curve of degree at most 6 which lies on a non-singular quadric surface in P3 (Step 1). Then we use the fact that deformations of Σ cover a codimension 1 subvariety of X to show "2 that the space H 0 (Σ, NΣ/X /tor) is non-zero (Step 2), and finally we compute the dimension of this space directly and get a contradiction (Step 3). Step 1. Let Y be an irreducible component of F (X) of dimension s ≥ 4. Let I ⊂ X × Y be the family of lines parametrized by Y, and denote by πX and πY the projections from I to X and Y respectively. Since Y is irreducible and the fibers of πY are lines, I is irreducible of dimension s + 1. Let r = dim πX (I). Note that by Lemma 4.5, X is not covered by lines, hence r ≤ 4. If r = 3, then the fiber of a general point in πX (I) has dimension s + 1 − 3 ≥ 2, so πX (I) is a linear subvariety of dimension 3 in X. Since X is smooth, it cannot contain a 3-dimensional linear subvariety by a similar argument as in the case of d = n = 4. We will argue that r = 4 also leads to a contradiction. Let X % = πX (I) and denote by p a general point of X % . If r = dim X % = 4, then there is at least a 1-parameter family of lines on X % which pass through p. Claim 4.7. No 2-parameter family of lines on X % pass through p. Proof. Assume on the contrary that there is a 2-parameter family of such lines. Then by Corollary 3.2, the degree of the cone of lines on X % passing through p is at most 2. Hence it is a cone over a surface of degree at most 2. Any quadric surface is covered by rational curves, so we get a contradiction by Theorem 2.1. Let Σ be an irreducible cone of lines on X % passing through p. From the above claim, together with Corollary 3.2, we conclude that Σ is a surface of degree at most 6, and it sits in the proper intersection of two hypersurfaces of degrees 2 and 3 in the embedded tangent plane to X at p. Let C be a hyperplane section of Σ which does not 15 pass through p. The above argument shows that C lies on a P3 and it is a component of the proper intersection of a quadric Q and a cubic T in P3 . Lemma 4.8. With the same notation as above, we have the following: (a) C is not rational. (b) The quadric Q is irreducible and non-singular. Proof. Part (a) follows from Theorem 2.1. As for (b), if Q is singular at a point q, then by Proposition 3.7, q is contained in T and it is a singular point of T . We can assume that q = (1; 0; 0; 0) and Q is the zero locus of the polynomial x21 − x2 x3 or the polynomial x1 x2 depending on whether it is irreducible or not. Hence T is given by a polynomial of the form x0 G1 + G2 , where G1 and G2 are homogeneous polynomials of degrees 2 and 3 in x1 , x2 , x3 . Therefore, the irreducible components of the intersection of Q and T are rational curves and in particular C is rational. This is not possible by part (a). Step 2. Recall that X % is a 4-dimensional subvariety of X and that there is a 1-parameter family of lines contained in X % passing through a general point. We denoted by p a general point of X % and by Σ an irreducible cone of lines on X % passing through p. We "2 use these assumptions to show that NΣ/X /tor has a nonzero global section. Proposition 4.9. Let X ⊂ Pn be a projective variety. Let P be a polynomial, and let G be a closed subscheme of HilbP (X) whose general point is a reduced subvariety of X of dimension m. If the subschemes of X corresponding to the points of G cover a k-dimensional subvariety of X, then for a general point [Y ] of G, we have "k−m H 0 (Y, ( NY /X )/tor) ,= 0. Proof. The proof is similar to the proof of Lemma 4.5. Let I ⊂ X × G be the family of subschemes of X parametrized by G, and denote by (q, [Y ]) a general point of I. Note that the Zariski tangent space to HilbP (X) at [Y ] is isomorphic to H 0 (Y, NY /X ) ([9], I.2.8), and we have the following commutative diagram: TI,(q,[Y ]) " TG,[Y ] ⊂ H 0 (Y, NY /X ) dp1 ! TX,q φ ψ " NY /X ! ⊗ κ(q), where p1 : I → X is the projection map, dp1 is its differential, and ψ is the dual of the map IY /X ⊗ κ(q) −→ mX,q /m2X,q . By our assumption, the image of p1 is k-dimensional and since (q, [Y ]) is a general point of I, the same is true for the image of the differential map dp1 . Therefore the image of 16 ψ◦dp1 is at least (k−m)-dimensional. The diagram is commutative, so the image of φ is "k−m at least (k−m)-dimensional. This implies that the map H 0 (Y, NY /X ) → NY /X ⊗ " k−m κ(q) is nonzero for a general point q in Y , hence H 0 (Y, NY /X / tor) ,= 0. "2 Corollary 4.10. In the situation of our problem, H 0 (Σ, NΣ/X /tor) ,= 0, and for a general point q ∈ Σ, there are global sections s1 and s2 of NΣ/X such that (s1 ∧s2 )(q) ,= 0. Proof. The statement is the consequence of the proposition above (and its proof for the second statement) with k = 4 and m = 2, along with the assumptions that our base field is uncountable and the cones of lines cover a 4-dimensional subvariety of X. Step 3. In this part, we try to compute h0 (Σ, using Corollary 4.10. "2 NΣ/X /tor) directly and get a contradiction Lemma 4.11. Assume Y ⊂ X are nonsingular varieties. Then for every subvariety Z of Y the sequence of normal sheaves 0 → NZ/Y → NZ/X → NY /X ⊗ OZ → 0 is exact. Proof. It is enough to show that the exact sequence of conormal sheaves IY /X ⊗ OZ → IZ/X ⊗ OZ → IZ/Y ⊗ OZ → 0 (9) is exact on the left too and splits locally. Since X and Y are nonsingular, we have an exact sequence of OY modules d 0 → IY /X ⊗ OY −→ Ω1X ⊗ OY → Ω1Y → 0, which splits locally. Therefore locally there exists a map s : Ω1X ⊗ OY → IY /X ⊗ OY such that s ◦ d = id. Now it is clear that the composition of the maps d s⊗id IZ/X ⊗ OZ −→ Ω1X ⊗ OZ −−−−→ IY /X ⊗ OZ splits sequence (9) locally. By the above lemma, the sequence of normal sheaves on Σ 0 → NΣ/X → NΣ/P6 → NX/P6 |Σ ∼ = OΣ (6) → 0 (10) is exact. Dividing the second exterior power of this sequence by torsions, we get the following short exact sequence 17 0→ #3 NΣ/X (−6)/tor → #3 NΣ/P6 (−6)/tor → #2 NΣ/X /tor → 0. (11) Also, by taking determinants of (10), we get the following isomorphism #4 #3 NΣ/P6 (−12)/tor ∼ = NΣ/X (−6)/tor. (12) By step 1, deg Σ ≤ 6 and it is a cone over a non-rational curve. Hence 3 ≤ deg Σ ≤ 6. The case deg Σ = 3 cannot happen because in this case C would be a curve of type (1, 2) on the non-singular quartic Q and any such curve is rational. So we have deg Σ " ≥ 4. We now analyze different cases separately. In each case, first we compute 2 h0 (Σ, NΣ/X /tor) by applying the long exact sequence of cohomology to sequence (11), and then we use Corollary 4.10 to get a contradiction. The case deg Σ = 4 : In this case, C is of type (1, 3) or (2, 2) as a divisor on Q. By Lemma 4.8, the former cannot happen because a curve of type (1, 3) on a quadric surface is rational. In the latter case, C is a complete intersection of two hyperplanes and two quadrics in P6 , and NΣ/P6 ∼ = OΣ (2)2 ⊕ OΣ (1)2 . "3 "3 6 (−6)) Hence we have H 1 (Σ, NΣ/X (−6)) = H 1 (Σ, OΣ (−6)) = 0, and H 0 (Σ, N "2 Σ/P = H 0 (Σ, OΣ (−1)2 ⊕ OΣ (−2)2 ) = 0. This implies, by (11), that H 0 (Σ, NΣ/X ) = 0, contradicting Corollary 4.10. The case deg Σ = 5 : In this case, C is a possibly singular divisor of type (2, 3) on Q. Let U be the complement of the vertex of Σ, and let π : U → C be the projection map. We show that: "3 (a) H 0 (U, NΣ/P6 (−6)|U ) = 0. (b) There is a sheaf of OΣ -modules G and a map #3 φ: NΣ/X (−6)/tor −→ G , such that φ is an isomorphism on U and H 1 (Σ, G ) = 0. Let us first show"how we can use map φ induces a "3these to get a contradiction. "The 2 2 map between Ext1 ( NΣ/X / tor, NΣ/X (−6)/ tor) and Ext1 ( NΣ/X / tor, G ). So we can extend sequence (11) to a commutative diagram 0 " "3 NΣ/X (−6)/tor " "3 NΣ/P6 (−6)/tor " "2 NΣ/X /tor ! G ! " "2 NΣ/X /tor φ 0 " " ψ G% 18 "0 " 0, and ψ is an isomorphism on U because φ is. Since H 1 (Σ, G ) = 0, every global section "2 of NΣ/X /tor is the image of a global section of G % and hence zero on U because "3 0 H (U, NΣ/P6 (−6)|U ) = 0. This contradicts Corollary 4.10. Proof of Part (a): Denote by P5 ⊂ P6 a hyperplane that contains C and does not pass through the vertex of Σ. Recall that π is the projection map from U to C, so the 5 normal sheaf of U in P6 is the pullback of the normal "3 sheaf of C in P under π. Hence to show the assertion, it is enough to show that NC/P5 (−6) has no nonzero global sections. Since C lies on the nonsingular quadric Q in P3 , NC/P3 and hence NC/P5 "3 ∨ are locally free sheaves, and NC/P5 ∼ = NC/P 5 ⊗ det NC/P5 . There is an exact sequence of conormal sheaves: ∨ ∨ ∨ 0 → OC (−2) ⊕ OC (−1)2 ∼ = NQ/P 5 |C → NC/P5 → NC/Q → 0. ∨ Since deg NC/Q = −C 2 = −12, from the sequence above we get deg(det NC/P5 ) = 32. Tensoring the above exact sequence with det NC/P5 ⊗ OC (−6), we get the following exact sequence 0 → det NC/P5 ⊗(OC (−8)⊕OC (−7)2 ) → #3 ∨ NC/P5 (−6) → det NC/P5 (−6)⊗NC/Q → 0. Since the right and left terms of this sequence are direct sums of negative degree line bundles, they have no nonzero global sections. So the middle term has no nonzero global sections either. Proof of Part (b): Since Σ lies on a P4 , by Lemma 4.11, the sequence of normal sheaves 0 → NΣ/P4 → NΣ/P6 → NP4 /P6 |Σ ∼ = OΣ2 (1) → 0 is exact. Taking determinants in this sequence, together with equation (12), we get #3 NΣ/X (−6)/tor ∼ = #2 NΣ/P4 (−10)/tor. Let IC be the ideal sheaf of C in P3 . Since H 0 (P3 , IC (3)) ≥ H 0 (P3 , OP3 (3)) − H 0 (C, OC (3)) = 6, the homogeneous ideal of C in P3 is generated by equations of the quadric Q and two cubics, and the intersection of each of the cubics with Q is the union of C and a line, so we get the following exact sequence of OP3 -modules OP3 (−4)2 → OP3 (−3)2 ⊕ OP3 (−2) → OP3 → OC → 0. (This sequence is exact on the left too, and it is the minimal free resolution of OC in P3 , but we won’t need that.) If I ⊂ k[x0 , . . . , x3 ] is the homogeneous ideal of C in P3 , 19 then the homogeneous ideal of Σ in P4 is generated by I in k[x0 , . . . , x4 ]. Hence the above sequence gives a similar exact sequence for Σ in P4 OP4 (−4)2 → OP4 (−3)2 ⊕ OP4 (−2) → IΣ → 0, (13) where IΣ is the ideal sheaf of Σ in P4 . Recall that from an exact sequence of modules T → M →"N → 0 over "2 a ring A, we get an exact sequence of second exterior powers 2 T ⊗M → M→ N → 0. So restricting sequence (13) to Σ, and then considering its second exterior power, we get the following exact sequence OΣ (−7)4 ⊕ OΣ (−6)2 → OΣ (−5)2 ⊕ OΣ (−6) → #2 IΣ /IΣ2 → 0. Finally, if we dualize this sequence and twist it with OΣ (−10), we get #2 η 0→( IΣ /IΣ2 )∨ ⊗ OΣ (−10) → OΣ (−5)2 ⊕ OΣ (−4) → OΣ (−3)4 ⊕ OΣ (−4)2 . (14) Denote by F the image of η. We claim that H 0 (Σ, F ) = 0 and H 1 (Σ, OΣ (−5)2 ⊕ OΣ (−4)) = 0. The first part is clear since F is a subsheaf of OΣ (−3)4 ⊕ OΣ (−4)2 which has no nonzero global sections. For the second part of the claim, notice that IC/Q ∼ = OQ (−2, −3), so H 1 (Q, IC/Q (m)) = 0 for every integer m ([8], III, Ex. 5.6). Therefore the map H 0 (P4 , OP4 (m)) −→ H 0 (C, OC (m)) is surjective for every m. The above map factors through the map H 0 (Σ, OΣ (m)) → H 0 (C, OC (m)) and hence this map is also surjective. So H 1 (Σ, OΣ (m)) = 0 for every integer m. " 2 Let G = ( IΣ /IΣ2 )∨ ⊗ OΣ (−10). It follows from the claim that H 1 (Σ, G ) = 0. Notice that there is a natural map #2 #2 NΣ/P4 (−10) = ( (IΣ /IΣ2 )∨ ) ⊗ OΣ (−10) −→ G , and since G is torsion free, the above map induces a map φ: #2 NΣ/P4 (−10)/tor → G . Finally, IU /IU2 is isomorphic to π ∗ (IC /IC2 ), and since the latter is locally free, φ is an isomorphism on U . The case deg Σ = 6 : In this case, Σ is the complete intersection of two hyperplanes, a quadric, and a cubic in P6 , and NΣ/P6 ∼ = OΣ (1)2 ⊕ OΣ (2) ⊕ OΣ (3). 20 "4 "3 Therefore H 1 (Σ, NΣ/P6 (−12)) = H 1 (Σ, OΣ (−5)) = 0, and h0 (Σ, NΣ/P6 (−6)) = ⊕2 0 h (Σ, OΣ ⊕ OΣ (−1) ⊕ OΣ (−2)) = 2. So from sequence (11) we get h0 (Σ, #2 NΣ/X ) = 2. Lemma 4.12. Every nonzero global section of NΣ/X has finitely many zeros. Proof. Let r ∈ H 0 (Σ, NΣ/X ) be a nonzero global section. Recall that det NΣ/X is isomorphic to OΣ (1) by (12). So by (11) we have an exact sequence: µ 0 → OΣ (−5) −→ OΣ2 ⊕ OΣ (−1) ⊕ OΣ (−2) → #2 NΣ/X → 0. (15) The map µ can be described in the following way. Let L1 , L2 , G, and K be homogeneous polynomials defining the two hyperplanes, the quadric, and the cubic in P6 whose intersection is Σ, and let P be a homogeneous polynomial defining X. There are homogeneous polynomials H1 , H2 , R, S such that P = H1 L1 + H2 L2 + R G + S K. Then the map µ in sequence (15) is given by (H1 ,H2 ,R,S) µ : OΣ (−5) −−−−−−−−−→ OΣ2 ⊕ OΣ (−1) ⊕ OΣ (−2). By the next lemma, there is s ∈ H 0 (Σ, NS/X ) such that r ∧s is a nonzero global section "2 of NΣ/X . Since H 1 (Σ, OΣ (−5)) = 0, r ∧ s is the image of some t in H 0 (Σ, OΣ2 ⊕ OΣ (−1) ⊕ OΣ (−2)). Without loss of generality, we can assume t = (1, 0, 0, 0). So if r(q) = 0 for a point q ∈ Σ, then H2 (q) = R(q) = S(q) = 0. The set {q ∈ Σ | H2 (q) = R(q) = S(q) = 0} is zero-dimensional, since otherwise there would be a point in the intersection of this set and the hypersurface H1 = 0 and this point would be a singular point of X. Therefore r has finitely many zeros. Lemma 4.13. For every nonzero r ∈ H 0 (Σ, NΣ/X ), there exists s ∈ H 0 (Σ, NΣ/X ) such that r ∧ s ,= 0. Proof. Let q be a general point of Σ. By Corollary 4.10, there are global sections s1 and s2 of NΣ/X such that (s1 ∧s2 )(q) ,= 0. So the image of the map H 0 (Σ, NΣ/X ) → NΣ/X ⊗ κ(q) is at least 2-dimensional. On the other hand, since q is a general point, r(q) ,= 0, so there exists s ∈ H 0 (Σ, NΣ/X ) such that s(q) and r(q) are linearly independent in NΣ/X ⊗ κ(q). Such a global section satisfies s ∧ r ,= 0. Let H be the Hilbert scheme of subschemes of X with the same Hilbert polynomial as Σ. Since dim X % = 4 and we chose Σ to be a general cone of lines on X % , the dimension of H at the point [Σ] is at least 4, so h0 (Σ, NΣ/X ) ≥ 4. Also, we showed "2 that h0 (Σ, NΣ/X ) = 2, so there are independent global sections r, s of NΣ/X such that r ∧ s = 0. Let C % be a hyperplane section of Σ which does not contain any zeros of 21 r. We have (r ∧s)|C ! = 0 and r|C ! is nowhere zero. Hence there exists a constant c such that r|C ! = c s|C ! . Therefore r − c s is a global section of NΣ/X with a 1-dimensional set of zeros and so by the previous argument it is the zero section. This contradicts our assumption that r and s are linearly independent. Remark 4.14. The above proof shows that for any smooth hypersurface of degree at least 6 in P6 , the Fano variety of lines is at most 3-dimensional, and as it was mentioned earlier, any Fermat hypersurfaces of degree at least 6 in P6 contains a 3-dimensional family of lines. References [1] W. Barth, A. Van De Ven, Fano-varieties of lines on hypersurfaces, Arch. Math. 31 (1978), 96-104. [2] R. Beheshti, Lines on Fano hypersurfaces, Thesis, M.I.T., 2003. [3] A. Collino, Lines on quartic threefolds, J. London Math. Soc. 19 (1979), no. 2, 257-262. [4] O. Debarre, Higher-dimensional algebraic geometry, Springer-Verlag, New York, 2001. [5] O. Debarre, L. Manivel, Sur la variété des espaces linéaires contenus dans une intersection complète, Math. Ann. 312 (1998), no. 3, 549-574. [6] P. Griffiths, J. Harris, Algebraic geometry and local differential geometry, Ann. Sci. Ec. Norm. Sup. 12 (1979), 355-432. [7] J. Harris, B. Mazur, R. Pandharipande, Hypersurfaces of low degree, Duke Math. J. 95 (1998), 125-160. [8] R. Hartshorne, Algebraic Geometry, Springer-Verlag, 1977. [9] J. Kollar, Rational curves on algebraic varieties, Springer-Verlag, Berlin, 1995. [10] J. Landsberg, Lines on algebraic varieties, J. Reine Angew. Math. 562 (2003) 1-3. [11] J. Landsberg, Griffiths-Harris rigidity of compact Hermitian symmetric spaces, to appear in J. Differential. Geom. Max-Planck-Institut für Mathematik, Bonn 53111, Germany. E-mail address : [email protected] 22